Regression: Python: Objects loaded from blend files have default matrices before depsgraph update #119348

New Issue

Closed

opened 2024-03-11 22:14:01 +01:00 by Vitalii-2 · 11 comments

Vitalii-2 commented

2024-03-11 22:14:01 +01:00

Seems like that should be a separate report. Indeed, when appending objects, the transforms won't be calculated until there's a depsgraph update.

Originally posted by @HooglyBoogly in #119250 (comment)

Blender Version
Broken: version: 4.2.0 Alpha, branch: main, commit date: 2024-03-10 18:36, hash: 703353b5dafc
Worked: (newest version of Blender that worked as expected)

Caused by 1c0f374ec3

Short description of error
When loading objects from blend files with python (bpy.data.libraries.load), loaded objects would have default value for world and local matrices instead of values stored in blend file.

Exact steps for others to reproduce the error
Place asset.blend and script.blend into same folder. Open script.blend and run the script. The script loads an object from asset.blend file and compares world matrix of an object before and after depsgraph update.
On 4.2.0 alpha this raises an exception as matrices don't match
On previous versions no exception is raised.

There is also a forward compatibility issue, as object loaded from blend files of 4.2 by older version of blender would also have default matrices before depsgraph update.

Seems like that should be a separate report. Indeed, when appending objects, the transforms won't be calculated until there's a depsgraph update. _Originally posted by @HooglyBoogly in https://projects.blender.org/blender/blender/issues/119250#issuecomment-1143815_ **Blender Version** Broken: version: 4.2.0 Alpha, branch: main, commit date: 2024-03-10 18:36, hash: `703353b5dafc` Worked: (newest version of Blender that worked as expected) Caused by 1c0f374ec3 **Short description of error** When loading objects from blend files with python (`bpy.data.libraries.load`), loaded objects would have default value for world and local matrices instead of values stored in blend file. **Exact steps for others to reproduce the error** Place asset.blend and script.blend into same folder. Open script.blend and run the script. The script loads an object from asset.blend file and compares world matrix of an object before and after depsgraph update. On 4.2.0 alpha this raises an exception as matrices don't match On previous versions no exception is raised. There is also a forward compatibility issue, as object loaded from blend files of 4.2 by older version of blender would also have default matrices before depsgraph update.

asset.blend

854 KiB

script.blend

949 KiB

Vitalii-2 added the

labels 2024-03-11 22:14:02 +01:00

Iliya Katushenock added the

labels 2024-03-12 10:17:52 +01:00

Philipp Oeser commented

2024-03-12 11:40:49 +01:00

Member

Can confirm

Can confirm

Philipp Oeser added

and removed

labels 2024-03-12 11:40:58 +01:00

Philipp Oeser added this to the 4.2 LTS milestone 2024-03-12 11:41:48 +01:00

Philipp Oeser added

and removed

labels 2024-03-12 11:42:25 +01:00

Philipp Oeser commented

2024-03-12 11:45:41 +01:00

Member

CC @HooglyBoogly (just noting it is only an issue in 4.2 -- so not relevant for next weeks release)

CC @HooglyBoogly (just noting it is only an issue in 4.2 -- so not relevant for next weeks release)

Philipp Oeser changed title from ~~Regrssion: Python: Objects loaded from blend files have default matrices before depsgraph update~~ to Regression: Python: Objects loaded from blend files have default matrices before depsgraph update

2024-03-12 12:27:19 +01:00

Brecht Van Lommel commented

2024-03-21 15:08:12 +01:00

Owner

What depends on these matrices being set immediately after appending?

In general, due to parenting, animation and constraints these matrices could already be mismatched with the actual object transform in the new blend file. Previously they just happened to be initialized to the value of the last evaluated frame in other blend file.

Depending on why and if this is important, possible options are:

Initialize based on location/rotation/scale without any parenting, animation or constraints
Restore old code to save them in the .blend file
Do nothing

What depends on these matrices being set immediately after appending? In general, due to parenting, animation and constraints these matrices could already be mismatched with the actual object transform in the new blend file. Previously they just happened to be initialized to the value of the last evaluated frame in other blend file. Depending on why and if this is important, possible options are: * Initialize based on location/rotation/scale without any parenting, animation or constraints * Restore old code to save them in the .blend file * Do nothing

Brecht Van Lommel removed the

label 2024-03-21 15:09:10 +01:00

Vitalii-2 commented

2024-03-21 15:51:18 +01:00

Author

Outside of python scripts that process object transforms before depsgraph update, the part of blender that relies on those matrices not being degenerate is bpy.ops.object.add_named() used to place objects from outliner into scene. Because appended object has world matrix with all 0, the the operator creates an object with no rotation and 0 for scale.

With animation transform of appended object not likely being valid, I'm voting on initialization to base transform. Maybe with parent if parent relationship got preserved on append?

Outside of python scripts that process object transforms before depsgraph update, the part of blender that relies on those matrices not being degenerate is `bpy.ops.object.add_named()` used to place objects from outliner into scene. Because appended object has world matrix with all 0, the the operator creates an object with no rotation and 0 for scale. With animation transform of appended object not likely being valid, I'm voting on initialization to base transform. Maybe with parent if parent relationship got preserved on append?

Roman Markov commented

2024-03-22 12:42:36 +01:00

The code that becomes broken for me is setting an object's matrix_parent_inverse from the parent object's matrix_world.inverted() as it is now all zeros and it gives ValueError: Matrix.invert(ed): matrix does not have an inverse or the default matrix that is incorrect.

Blender 4

Blender 4.2

The code that becomes broken for me is setting an object's `matrix_parent_inverse` from the parent object's `matrix_world.inverted()` as it is now all zeros and it gives `ValueError: Matrix.invert(ed): matrix does not have an inverse` or the default matrix that is incorrect. Blender 4 ![image](/attachments/1ef0d983-ddaf-479a-a273-713b6931bdc8) Blender 4.2 ![image](/attachments/71825183-38a8-435a-8849-9c0b129b107c)

286 KiB

256 KiB

Roman Markov commented

2024-03-22 12:58:56 +01:00

place objects from outliner into scene [...] object has world matrix with all 0

With the 4.2.0 Alpha, branch: main, commit date: 2024-03-21 22:18, hash: c61ecf1f4094 it is now the default one. So if the original scale of the object was 3 you will get 1.

> place objects from outliner into scene [...] object has world matrix with all 0 With the `4.2.0 Alpha, branch: main, commit date: 2024-03-21 22:18, hash: c61ecf1f4094` it is now the default one. So if the original scale of the object was 3 you will get 1.

Hans Goudey commented

2024-04-04 19:00:40 +02:00

Member

Given the options outlined by Brecht above, I'd argue for "Do nothing". The matrices when appending the object before depsgraph evaluation should be identity matrices. Some reasoning:

Arguably the previous behavior was never purposeful, it just happened because all runtime data was written to files.
It would be good to avoid adding object-specific code to data-block append/link
Initializing the matrices based on loc/rot/scale may have some performance cost that we wouldn't be able to fix
The new behavior clarifies the use and implementation of the structs, and is more self-consistent.

Given the options outlined by Brecht above, I'd argue for "Do nothing". The matrices when appending the object before depsgraph evaluation should be identity matrices. Some reasoning: - Arguably the previous behavior was never purposeful, it just happened because all runtime data was written to files. - It would be good to avoid adding object-specific code to data-block append/link - Initializing the matrices based on loc/rot/scale may have some performance cost that we wouldn't be able to fix - The new behavior clarifies the use and implementation of the structs, and is more self-consistent.

Bastien Montagne commented

2024-04-13 19:11:46 +02:00

Owner

Agree with @HooglyBoogly for current situation. If scripts depend on runtime evaluated data, then they should ensure that the data is evaluated.

We do have a forward compatibility issue here though, in that when opening a 4.2 blendfile in Blender 4.1 or older, these matrices will be initialized to null (and not identity) by the DNA reading code.

The only way to fix this issue I can think of currently is to keep writing dummy identity matrices in Object data for the time being (until we reach 5.0), to avoid getting invalid transform matrices in older versions... Not sure how critical this fix is though.

Agree with @HooglyBoogly for current situation. If scripts depend on runtime evaluated data, then they should ensure that the data is evaluated. We do have a forward compatibility issue here though, in that when opening a 4.2 blendfile in Blender 4.1 or older, these matrices will be initialized to null (and _not_ identity) by the DNA reading code. The only way to fix this issue I can think of currently is to keep writing dummy identity matrices in Object data for the time being (until we reach 5.0), to avoid getting invalid transform matrices in older versions... Not sure how critical this fix is though.

Hans Goudey commented

2024-04-13 19:38:20 +02:00

Member

We do have a forward compatibility issue here though, in that when opening a 4.2 blendfile in Blender 4.1 or older, these matrices will be initialized to null (and not identity) by the DNA reading code.

I think we can handle that with versioning code that checks if the file is from the future and initializes identity matrices in that case. We can patch 4.1 and 3.6 anyway. Given the non-critical aspect of the problem (data is properly filled in as soon as the object is evaluated), I'd think that should be enough.

>We do have a forward compatibility issue here though, in that when opening a 4.2 blendfile in Blender 4.1 or older, these matrices will be initialized to null (and not identity) by the DNA reading code. I think we can handle that with versioning code that checks if the file is from the future and initializes identity matrices in that case. We can patch 4.1 and 3.6 anyway. Given the non-critical aspect of the problem (data is properly filled in as soon as the object is evaluated), I'd think that should be enough.

Bastien Montagne modified the milestone from 4.2 LTS to 4.1

2024-04-15 11:31:54 +02:00

Bastien Montagne commented

2024-04-15 11:32:41 +02:00

Owner

@HooglyBoogly if this is to be fixed that way, then we need a PR for 4.1.1 within the coming hours. ;)

@HooglyBoogly if this is to be fixed that way, then we need a PR for 4.1.1 within the coming hours. ;)

Hans Goudey referenced this issue from a commit

2024-04-15 15:39:32 +02:00

Fix: Set Object runtime matrices to identity when loading 4.2 files

Hans Goudey commented

2024-04-15 16:37:17 +02:00

Member

4.1 has been changed to set the identity matrix on file load. There's nothing more to do here, so I'll close this task.

4.1 has been changed to set the identity matrix on file load. There's nothing more to do here, so I'll close this task.

Hans Goudey closed this issue

2024-04-15 16:37:21 +02:00

Blender Bot added

and removed

labels 2024-04-15 16:37:24 +02:00

Sign in to join this conversation.

No Label

Animation & Rigging

Asset Browser Project Overview

Automated Testing

Blender Asset Bundle

Dependency Graph

Development Management

EEVEE & Viewport

Images & Movies

Motion Tracking

Nodes & Physics

Pipeline, Assets & IO

Platforms, Builds & Tests

Render & Cycles

Render Pipeline

Sculpt, Paint & Texture

Video Sequencer

Virtual Reality

Blender 2.8 Project

Milestone 1: Basic, Local Asset Browser

Good First Issue

Animation & Rigging

Development Management

EEVEE & Viewport

Nodes & Physics

Pipeline, Assets & IO

Platforms, Builds & Tests

Render & Cycles

Sculpt, Paint & Texture

Needs Info from Developers

Needs Information from User

No Milestone

No project

No Assignees

6 Participants

Notifications

Due Date

The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: blender/blender#119348

No description provided.